Envisioning explanations – the art in science

Forthcoming in Interdisciplinary Science Reviews, vol. 29, 2004, Special Issue on Scientific Illustration.
Envisioning explanations – the art in science
DAVID C. GOODING
Science Studies Centre, Department of Psychology, University of Bath, Bath, U.K.
Visualization involves making and manipulating images that convey novel phenomena, ideas and
meanings. It is central to the intellectual objectives of almost every area of science. It is therefore
surprising that we still lack an understanding of how visual thinking works. Examples from different
sciences illustrate how scientists interpret and explain phenomena by actively manipulating those
pattern-like features of the world that lend themselves to structural representation, and vice versa. The
ability to move between two, three and four dimensions structures both imagination and analysis,
enabling scientists to vary the complexity of each visualization as its function changes during the
process of discovery and communication. These findings suggest that there is a widely used visual
method that relates human cognitive abilities to imaging practices in the sciences and the arts.
Just as Leonardo daVinci attempted to capture turbulent water in a famous series of sketches, scientists and
their illustrators have endeavoured to image process.1 But printed images cannot display process directly to
our senses as animations and simulations do. Cinematography, multimedia and virtual reality technologies
can appeal directly to the multi-sensory character of popular culture and are displacing the ordered
unfolding of traditional narrative and discourse. A recent conference on “Visual Knowledges” addressed
the proposition that “our culture is currently, in the wake of the electronic revolution, undergoing a shift in
which the visual medium, traditionally playing a secondary role as the illustration of text, is becoming the
dominant medium of thought”.2 The trend away from narrative, argument and intellectual comprehension
toward unstructured, experiential modes of apprehending is reflected in the performative and experiential
style of many approaches to raising public awareness of science. Are promoters of science aping popular
culture? Far from it. The sensory technologies that drive popular culture draw on the same features of
human experience that the sciences have always exploited. These include our abilities to recognise patterns,
translate different types of sensory experience into words and pictures and integrate these into hypotheses,
models and theories about processes. Stimulating our experiential abilities has of course made film,
multimedia and virtual-reality technologies more appealing than the narratives and static images of printed
books. Yet the sciences have always had to push at the limitations of printed media because, like
technologies of animation and virtual reality, science is mostly about processes we cannot experience. Facts
and data are expressed visually (as patterns), numerically (as datasets), or verbally (by declarative sentences)
but these static descriptions are just ways of apprehending a complex and dynamic world of change. One of
the main motives of scientific inquiry is to deliver a general, intellectual understanding of the processes that
produce our experiences.3
Visualization – making and manipulating images that convey novel phenomena, ideas and
meanings – is central to this intellectual objective in almost every area of science. It is therefore surprising
that we still lack an understanding of how visual thinking works. Case-studies of image-making in science
display great variety in visual imagery and representational practices.4 Variety is all-important in historical
and cultural studies: it identifies emerging genres, practices and consensus about representational
conventions and what is depicted by them. Each style of representation, instrumentation, or argumentation
defines a distinct culture. Historians have shown how modern science emerged by forging a new identity as
a cultural activity distinct from the visual arts, crafts and technology.5 Diversification differentiates
disciplines and even points to important differences within a discipline, such as the emergence of two
distinct ways of practising high energy physics according to their conflicting views about the epistemic
status of images and measured quantities.6 These case studies treat diversity in the manner of natural
history: there is no expectation of anything else.
Scientists take a less pluralist view, as do many philosophers and psychologists. This usually
characterises the sciences in terms of methods for producing results and standards for evaluating them.
Robert May described the essence of all scientific method this way: "the world is complicated but not all the
complexities are important all the time" so that "the real trick is to intuit ... the subset of things that are
important, and on this basis to formulate a tentative understanding in unambiguous terms ". You then
"pursue where that tentative understanding leads, test it against the facts, and usually circle back to refine
the original assumptions". He added that "this basic process holds whether you are trying to understand
superconductivity, or the causes and consequences of biological diversity, or how best to translate Britain's
excellent science base into industrial strength".7 Philosophers cast science more narrowly as a rational
process of reasoning about facts according to logical or statistical rules of inference.8 Many cognitive
psychologists also assume logical or statistical models of human inference.9 Evolutionary explanations
purport to show that scientific methods depend on cognitive capabilities closely linked to the survival and
evolutionary supremacy of humans and which, it is argued, have reached their most developed form in the
methods of science.10
These positions claim superiority for scientific ways of knowing. However the idea of a common
method smacks of epistemological privilege which most scholars in the humanities and in science studies are
keen to deny to any facet of culture, especially science. The nature and status of science remain as contested
as ever. So is science simply diverse, a panoply of local, culturally situated practices? Or is there a human
factor common to many sciences and perhaps to other activities as well? Martin Kemp remarks that “it is
inevitably the differences that most strike us, particularly as historians of culture” but he also observes that
“beneath the varied surface run some constant currents in our human quest for visual understanding. The
most enduring of these currents is our propensity to articulate acts of seeing through “structural
intuitions”.11 This phrase connotes the structures both of “inner intuitive” processes and of “external
features whose structures are being intuited”. I will elaborate both aspects of this notion with examples that
illustrate another aspect of variability. Scientists articulate their intuitions by making representations whose
complexity can be varied. They actively manage complexity by manipulating those pattern-like features of
the world that lend themselves to structural representation, and vice versa. The complexity of a visualization
varies according to its function at each stage in the process of discovery and communication. The cases I
have studied suggest a model of the process of articulating visual intuitions as visual constructs. This relates
human cognitive strengths and weaknesses to the diversity and richness of imaging practices in a way that
linguistic and logical models of scientific inference have failed to do.12
–– Envisioning Explanations, page 2 of 16 ––
Managing complexity - visual method in science
Visual images offer convenient windows into activities that are personal and imaginative as well as the more
visible, conventional modes of communication of the public domain. Each case is unique, yet each shares
common features and similar patterns of movement between the personal (or private) domain of mental
images and the public domain of communicated concepts. These patterns and features define a visual
method in the play of cognition as the locus of reasoning and culture as the locus of construction of images,
narratives meanings, arguments and objects. This method has its counterparts in artistic technique and its
procedures are often embodied in new technologies.
In the visual and graphic arts images and objects are self-sufficient as modes of expression and
carriers of meaning. In science the role of images and artefacts changes: at first conveying a tentative
understanding but later aiding communication and the construction of arguments. These generative and
communicative roles usually fall away as new verbal and symbolic expressions become established. The
declamatory caption given to the first published image of the double helical structure of DNA is a typical
example. Although structural modelling had been essential to Watson and Crick’s solution their first paper
does not include an image of the final model. Instead there is a diagram whose caption indicates that the
image is a purely an aid to understanding the written text.13 The role and status of images follows its own
pattern of change in each case. When we look for common features in these patterns of visual inference we
can see how apparently disparate elements of scientific practice -- words, sketches, photographs, models,
instruments, and other traces and products of scientific activity -- work together in making and
communicating new knowledge. This comparative approach can take us beyond the surface history of
imagery in science, allowing the visual proclivities of scientists to show how their creativity is like that of
other humans, and also how it differs.
The method described here foregrounds visual images in order to indicate how they relate to
narrative forms of expressing thoughts and arguments and to visual technologies. It does this by relating the
complexity of an image to the purpose that it has at a particular time. There is a dialectical play of simple
and complex images, as scientists move from interpreting novel or unfamiliar information through
hypotheses expressing its possible meanings, to arguments that situate the information as evidence for an
agreed explanation. This variety points to an underlying, cognitive aspect of visual thinking: humans can do
some things quickly and automatically (such as seeing and comparing patterns) while others, such as
imagining the changing shape of an object, require mental effort. Variable complexity also explains an
important group of technologies – the artefacts and procedures devised to manage the difficulty of using
visualization and to communicate its emergent meanings.
Why should there be a dialectic rather than a linear movement from, say, simple representations to
more complex ones? Most models of method imply movement from complex to simple, such as the ‘single
pass’ models of discovery favoured by philosophers and some models of practice in science education and
science communication. These apply to finished work, where in writing up scientists edit out everything
that success has shown to be superfluous.14 But cutting-edge research encounters the world as a complex
and often unruly affair. Abstracting from what is too complex to interpret, communicate or reason about
makes new kinds of thinking possible. A simple object can be manipulated more readily in thought, by
hand or by computation and it can draw on the resources of a established, informal modes of
communication. Michael Faraday is typical of the scientists described in this paper. He wrestled continually
with the related difficulties of interpreting and communicating novel phenomena. Between 1821 and 1831 he
–– Envisioning Explanations, page 3 of 16 ––
adapted the discourse of ‘lines’ and ‘curves’ used in contexts as diverse as electromagnetism, geometry and
geomagnetism, developing it into a powerful visual language for an electromagnetic theory.15 But simple
representations leave out features that later prove to be important and by 1832 Faraday's intimate
knowledge of the interactions of electricity, magnetism and motive force exceeded the descriptive
capabilities of his earlier attempts.16 So he devised a structural model which expresses the observed
relationships between these three ‘powers’ by forces acting not between magnets, charges or other objects,
but between lines of electric, magnetic and motive action. His sketch of this abstraction (see figure 1) is
accompanied by verbal instructions that ‘animate’ the structure as a process model.17
“The mutual relation of electricity, magnetism and motion may
be represented by three lines at right angles to each other, any
one of which may represent any one of these points and the
other two lines the other points. Then if electricity be determine
in one line and magnetism in another, motion will occur in the
third. Or if magnetism be determined first then motion will
produce electricity or electricity motion. Or if motion be the
first point determined, Magnetism will evolve electricity or
electricity magnetism”.
FIGURE 1. Faraday’s sketch of March 1832. Together with the text which shows how to animate the image, this
expresses Faraday’s discovery of the three-way mutuality of electric, magnetic and motive actions. From:
Faraday’s Diary, vol. 1, p. 425 (see note 17)
This succinct visual-verbal statement is typical of how simple descriptions are made to convey
complex experience or to postulate new relationships. The more complex a representation is, the more
power it has to integrate different kinds of information derived from various sources. Increased complexity
and information density make it more difficult to think with in a consistent and general manner, leading to a
new cycle of abstraction. I shall illustrate this by examples from three areas – the reanimation of fossils,
reconstructive visualization in hepatology and stereographic projection in early X-ray crystallography.
These suggest wide use of a distinctive visual method of thinking in the sciences.
Re-animating extinct organisms
Flattened organisms preserved as fossils in sedimentary rock offer glimpses into the history of life. To
contribute to the story each fossil must be reconstructed as a plausible model of an organism. To be
plausible it must be identified as an instance of a known – or possibly a new – species. How do
paleontologists get from hints of structure in pieces of rock to plausible working models of organisms? They
reason with sketches, diagrams and drawings and manipulate physical models to produce new images. The
fossils of the Burgess shale are remarkable in that many are complete and even traces of soft parts and
internal organs are preserved.18 Fixing the identity of any single impression is complicated by several
factors. Shales split along cleavage planes to reveal a fossil imprint, so 'part' and 'counterpart' (the matching
positive and negative imprints) are often dispersed by geological process and human disturbance or into
different geological collections. The cleavage planes ‘cut’ the flattened organisms at many different angles.
Given these and many other difficulties, fossil imprints are interpreted by analogy to the morphology of
modern counterparts. However most Burgess organisms became extinct and have no modern analogues.
–– Envisioning Explanations, page 4 of 16 ––
Reconstructing fossil imprints involves creating images of the traces of an organism from the rock
using a camera lucida. These are diagrams of features whose function and relation to the anatomy remains
conjectural. The 2-D diagrams are provisional interpretations of the more complex phenomenology of the
‘raw’ fossil imprints. They are labelled to indicate structures that could represent a plausible anatomy.19 Of
course, expert knowledge about possibilities influences the selection of features from the diagram. A
reconstruction is considered plausible if it produces a model of an organism that could have lived.
FIGURE 2. Camera lucida drawings of two specimens of Leanchoilia superlata, from Bruton, D. L. & Whittington,
H. B. (1983), “Emeraldella and Leanchoilia, two arthropods from the Burgess Shale, Middle Cambrian, British
Columbia, Phil. Trans. Roy. Soc, B300: 553-585, figures 96 & 97 (reprinted by permission of the Royal Society).
A morphological interpretation of the flat photograph of a fossil is guided by the camera lucida
diagram generated from the photograph. The diagram mediates between flattened features extracted from
the photographic image and an emerging, 3-dimensional model of a possible organism (see figure 3). To be
identified as a section of a particular organism, each impression must be associated –– via its camera lucida
diagram –– with other impressions that may be images of the same animal captured in different orientations,
or images of completely different animals. Every fossil impression must be re-imaged mentally as if from
several points of view since “failure to recognize the effects of variable orientation may lead to misidentification”.20 Mental images are sketched or modelled as sections. Each one of these images is evaluated
as a possible match for other known impressions. One investigator (Conway Morris) reports having drawn
specimens found in various orientations, then passing countless hours “rotating the damned thing in my
mind” from the position of one drawing to the different angle of another, until every specimen could be
moved without contradiction from one stance to the next”.21 These painstaking, image-based searches for
visual matches are essential to reduce the number of possible matches.
Conway Morris' phrase "without contradiction" hints at informal, implicit rules for transforming and
comparing mental images. In science (as in art) informal procedures are often replaced by techniques. By
projecting the changing shapes of shadows cast by sections of models the laborious image-based ‘thought
experiments’ can be replaced by systematic, model based transformations. Once a set of patterns from
–– Envisioning Explanations, page 5 of 16 ––
FIGURE 3. Reconstruction of the arthropod Sidneyia inexpectans as a 3-D model built in sections derived from
camera lucida drawings such as those in figure 2, from Bruton, D. L. (1981), “The arthropod Sidneyia
inexpectans, Middle Cambrian, Burgess Shale, British Columbia, Phil. Trans. R. Soc., B295: 619-656, figure 107
(reprinted by permission of the Royal Society).
different imprints is associated with an organism, each image can be treated as a section of a 3-D structure
representing the morphology of that organism. Where diagrammed imprints of the organism cannot be
made to fit sections of known animals, a new 3-D structure is conjectured.
At this stage the process of identifying imprints already involves a dialectical play between 2-D
images of patterns, 2-D images of 3-D structures, actual 3-D structures and their 2-D shadow-projections,
and the animation of these structures as living organisms. The 3-D visual model generated from the set of
matched or correlated 2-D impressions integrates several different types of information. It must be
consistent in a cognitive sense with a large set of mental and physical images and be consistent in a conceptual
sense with expert knowledge about the morphology of similar organisms. This must in turn be compatible
with other kinds of knowledge about life processes, such as the animal’s ecology (food sources, predators
and so on). We could say that the 3-D model is repeatedly tested in two ways. The first involves showing
that a living organism having the postulated structure would, when buried in sediments and subjected to
known geological, chemical and mineralization processes, produce the physical impression(s) from which
the original diagrams were abstracted. Thus the camera lucida diagram is used both as an interpretative
source of possible models and as an empirical check on animations of their structural projections. The
second type of test involves creating a life-process (or 4-D) model that integrates information from
disciplines such as ecology, anatomy, physiology, and evolutionary biology so as to evaluate the 3-D
structure as a model of such an animal. This illustrates how structure models mediate between images of
patterns and theories about process.
The interpretations of a few scientists must be evaluated by a wider group of experts. For use in the
public forum of negotiation and argument the photographs are retouched to highlight features selected as
significant by a provisional identification. The diagrams are also redrawn as ‘reconstructions’ that include
features that can be inferred though they are missing from the original images. These are shown in
drawings of the organism from various views (see figure 4).22 Each photographic image is paired with a
labelled camera lucida diagram and text. This shows readers how to interpret the 2-D images in terms of the
3-D reconstruction described in the paper. Using these shared objects of discussion readers can re-enact the
play of images of flattened layers, diagrams and structures.23
–– Envisioning Explanations, page 6 of 16 ––
A
B
C
FIGURE 4. A selection of images from the reconstruction of the arthropod Sanctacaris uncata. 4A: Photograph
of a counterpart (dry specimen, dorsal view). 4B: Labelled drawing based on the photograph. 4C:
Reconstruction drawing based on several fossil ‘sections’. From Briggs, D. & Collins, D. (1988). A middle
Cambrian Chelicerate from Mount Stephen, British Columbia, Paleontology, 31: 779-798, Text-Fig 1B; Plate 71
fig. 3; Text-Fig. 6. Copyright The Paleontological Association.
How Pattern and Structure elucidate Process
This example illustrates some key features of visual thinking also found in other cases. First, there is a
dialectical play of personal and public representations.24 Second, diagrams select and highlight certain
features of the ‘raw’ (un-interpreted) imprint. Third, a technology (the camera lucida) is used to produce
abstractions which scientists then work with, using techniques devised to bypass difficult mental tasks such
as working out 2-D projections of complex 3-D structures.25 Fourth: the selected features are used to
construct a 3-D model capable of integrating information from diverse knowledge domains. Fifth:
information-rich, integrated representations are developed into process models that explain both the original
2-D imprints and the 3-D anatomical features they are now shown to display.26 Sixth: the complex models
are evaluated by abstracting simpler visualizations from them and comparing these to the source patterns.
This set of strategies is motivated by the goal of most scientific disciplines – starting with complex
experience produced by some process, to explain visualized experience of patterns and structures in terms of
process.27 In paleontology the physiology and life process cannot be presented directly to the mind’s eye.
They are reconstructed verbally and pictorially by an informed, guided use of imagination, like Faraday’s
verbal instructions for animating his structural model of electricity, magnetism and motive force (see figure
1). Similarly, a geological block diagram is a 2-D (printed) image of a 3-D (structural) model designed to
explain a lengthy geological process. In block diagrams texture and colour identify different types and states
of rock. Signs such as arrows indicate the need to iterate versions of the imaged model through a series of
steps. Thus time (a non-spatial dimension) is represented by a symbol which is read as an instruction to
follow a procedure. Mathematical equations and operators such as integral signs are ‘read’ in a similar way.
In visualizations such as data plots time is represented spatially rather than by symbols. I refer to images
–– Envisioning Explanations, page 7 of 16 ––
and composite visual-symbolic representations that include indicators of process as four-dimensional. By
analogy un-interpreted ‘sources’ such as fossil imprints or ‘raw’ numerical data have zero dimensionality.
The ability to move easily between visualizations of differing complexity is a very important feature
of scientific thinking. Why should this be? In our examples each transformation increases or decreases the
complexity of the new visualization it produces in response to the constraints and opportunities posed at the
point at which it is invoked. If the information content of representations is related to the number of
dimensions then more dimensions imply greater capacity to convey meaningful information. More
dimensions indicate that greater mental or computational effort is needed to manipulate this information
consistently and to communicate what it means. This problem drives the invention of techniques and
technologies that simplify and regulate moves between images and models of different dimensionality.
In the diagram in figure 5 each of the four nodes denotes representations having different
dimensions while the arrows denote manipulations that transform simpler into complex, and vice versa. At
node A is the unresolved, complex phenomenology (of zero dimensionality); at B we have 2-D images
(sketches or diagrams) that abstract features from A and provide resources for structural models at C.
Structures can be varied by transformations such as mental rotation or physical projection of three
dimensions onto two. Visualizations at nodes B and C also provide a framework for integrating other kinds
of knowledge into a 4-D process model at node D. The structure and process models are validated if source
information as imaged at B can be re-derived from C or from D, along with new patterns or
C -- STRUCTURE
B – PATTERN
D -- PROCESS
A -- SOURCE
FIGURE 5. A depiction of visual inference showing the four main transformations. Moves from pattern to
structure (arc BC) and from structure to process (DC) are generative and expand information content. Moves
from unresolved phenomena to a pattern (arc AB), from process to structure or from structure to pattern (CB,
DC) are also generative but reduce the information content of an image.
structures. This diagram depicts some of the images, objects and processes that contribute to the variety of
visual imaginings in science. Pattern matching requires representations having two dimensions whereas
informal, creative inferences generate representations having different dimensionality.
The pattern of images, objects and inferences is unique to each case, depending on contingent factors
which influence stylistic and other features of the representations. Yet the types of transformations shown in
figure 4 occur in many cases. The scheme therefore offers a basis for the systematic comparison of episodes
of scientific discovery. It shows that visualizations bear a distinct, inferential relationship to each other,
relating each occurrence of an image or model to the larger context of observation, experimentation,
modelling, dissemination and argument in which it is used.
–– Envisioning Explanations, page 8 of 16 ––
Reconstructive visualization of vascular structures
The liver regulates the level of glucose in the blood, given constant fluctuations in both the nutritional
supply and demand. Stable levels are maintained by opposing systems of enzymes that act to release
glucose or to reduce its uptake into the blood. The enzyme chemistry that balances supply and use of blood
sugars is well understood. But for some mammalian livers there is a long-standing puzzle about the
physiology that supports the biochemical function. In some mammals dissection does not support structural
hypotheses because structures which must exist cannot be observed by the usual method of visually
inspectiing sections.28 As in the previous example the problem is to find a physiologically plausible structure
but here visual inspection of tissue sections provides few clues. Teutsch and his collaborators solved this by
envisioning the structures supporting the process from a spatially distributed set of measurements. This
produced a functional model of the elusive primary- and secondary vascular structures.
A
B
Figure 6. Maps that organise numerical data. Microdissection map (6A) showing subdivision of a liver lobule
cross-section into eight adjacent sectors. Histochemically stained sections (6B) are used to guide this
subdivision. The abbreviated labels indicate sections of entities such as the central venule (labelled ‘CV’)
whose structure will emerge (approximately) perpendicular to the plane of the imaged section as data is
processed and imaged. (From Teutsch, H. (1988). Regionality of Glucose-6-Phosphate Hydrolysis in the Liver
Lobule of the Rat, Hepatology, 8: 311-317, figures 1 and 2). Copyright J. Wiley & Sons, Inc.
To develop an empirical method that could also decide between different models of the relationship
between liver function and its structure, they organised a numerical dataset measuring biochemical
processes in dissected livers first into two-dimensional and then three-dimensional arrays. Sections are
mapped and imaged as in figure 6. To achieve a spatial separation of the opposed catalytic actions of the
enzymes, they made a closer study of rates of glucose uptake and release in cryosections of many rat livers.
Further visualizations plot gradients of activity rather than discrete values (figure 7). Increasing the quantity
of data improved the 'resolution' of the plots, making it possible to infer sub-modular structures consisting
of populations of cells.
Teutsch demonstrated modularity and vascular structure by constructing 3-D images from virtual
'stacks' of images of very thin cryosections.29 A sample from a much larger set of drawings of these is shown
in figure 7b. A particular structure of the modules and of the ducts that supply and drain them was
produced as a set of 3D drawings (figure7c). The final result of the reconstructive method is a 4-D or process
model that provides the physiological explanation of the source images of visualized data. This was
validated with 3-D image printing technology.
30
The 3-D images guided further construction of colour-
coded models in paste and plastic (see figure 8). Actual construction is needed to work out particular
–– Envisioning Explanations, page 9 of 16 ––
A
B
C
Figure 7A. Plots of the distribution of enzyme activity over cross-sectional areas of a primary modular structure
of rat liver, reconstructed as 3-D surfaces. The central vertical line denotes the location of the central venule
(labelled ‘CV’ in figure 6B). Reproduced, by permission of the Histochemical Society, from Teutsch, H., Altemus,
S., Gerlach-Arbeiter, S. & Kyander-Teutsch, T. “Distribution of 3-Hydroxybutyrate Dehydrogenase in Primary
Lobules of Rat Liver”, Journal of Histochemistry and Cytochemistry, 40: 213-219, 1992, figure 4. Fig. 7B.
Drawings of some of the 146 cross-sections of a secondary unit showing sections of the portal tracts and septa,
from which a 3-D structure (shown at right, Fig. 7C) was constructed (From Teutsch, H., E., Shuerfeld, D. and
Groezinger, E. (1999). Three-Dimensional Reconstruction of Parenchymal Units in the Liver of the Rat,
Hepatology, 29, 494-505, figs. 3 & 6. Copyright J. Wiley & Sons, Inc.
features, such as how modular structures incorporate portal and venal ducts. Teutsch presented this method
of reconstructive visualization as demonstrating complex vascular structures that are now considered as
facts established by reconstruction. Both their meaning and their factual status emerge from this method of
visual modelling.
Reconstructive visualization combines veridical images, sketches, schematic images, 2-D and 3-D
plots of numerical data, 3-D stacks of 2-D plots and physical models. 3-D printing technology automates the
process of assembling a 3-D structure from a large set of data-generated 2-D section plots. Just as drawing
directs attention to surface features, physical modelling checks the efficacy of a structure with regard to
function, as we saw in the case of the Burgess shale fauna, or to phenomenology as shown by the next
example, X-ray crystallography. This conjunction of images, models, instruments and theories shows how
diverse kinds of artefact extend the range of what can be visualized in a creative yet systematic way.
–– Envisioning Explanations, page 10 of 16 ––
FIGURE 8. Images of a model that is also an artistic reconstruction comparable to Sanctacaris in figure 4c. A is
the front view of a of a secondary parenchymal unit (left) and B shows the central venular tree that drains it. In
the original colour coding is used to differentiate the primary units of which the whole is composed. This
coding is lost in monochrome reproductions. From Teutsch, H., E., Shuerfeld, D. and Groezinger, E. (1999).
Three-Dimensional Reconstruction of Parenchymal Units in the Liver of the Rat, Hepatology, 29, 494-505. figure
4. Copyright J. Wiley & Sons, Inc.
Stereographic projection of crystal structure
In 1912 von Laue published images formed on a photographic plate by X-rays passing through a crystal (see
figure 9a). He surmised that if the wavelength of the X-rays is of the same order of magnitude as the
distance between atoms in the crystal, then the crystal will behave like a diffraction grating. Each plane in
the crystal deflects the incident ray at a slightly different angle, splitting it into interfering rays. This process
is happening in three dimensions rather than two. In the same year W. L. Bragg published the first of his
papers on crystal structure inspired by “the intricate geometrical patterns” produced by von Laue. These
papers established a method of inferring crystal structure from features of the pattern of spots produced by
the X-rays.31 These features include symmetry, actual position, the size of each spot and – it would soon
emerge – points at which spots do not appear. The variable size of the spots led Bragg to surmise that –
contrary to von Laue – any wavelength can be dispersed by a crystal but that as each successive plane
deflects the rays, there is interference between rays only when some relationship between wavelength and
inter-atomic distance is satisfied (see figure 9).
Symmetry is to be expected if crystals behave like three-dimensional diffraction gratings. The
position and magnitude of each spot and positions where spots are absent are clues to the relative positions
of the planes in which atoms of the crystal might lie. The spots appear where there is positive interference
between the rays. So a spot indicates an intersection between cones that represent the maximum dispersion
of each set of rays. An array of spots thus defines sets of intersections that in turn define cones whose
origins should be in the crystalline plane in which the atoms that are deflecting the beams lies. Bragg
transformed X-ray photographs into diagrams by pricking through the photos onto paper on which he then
marked the size of the spots. He then drew ellipses through them (see figure 9c). The ellipses represent
sections of the conical pencils of rays that converge inside the crystal at a point where the atoms lie. Bragg
realized that by varying the distance of the plate from the source he could also work out the geometry of the
action of the crystal on the X-ray beam. Similarly, varying the angle of incidence by moving the crystal
–– Envisioning Explanations, page 11 of 16 ––
produced displacements and distortions which could, when actively ‘read’ in terms of this geometry,
provide additional information about the relative positions of the crystal planes.32
A
B
C
FIGURE 9. From patterns to molecular structure. 9A. Photograph of pattern produced by diffraction of X-rays
by a crystal lattice. From Bragg, 1913, figure 3 (see note 31). 9B. Excerpt from a notebook entry showing the
pattern and size of spots being abstracted from an X-ray image (Courtesy of the Royal Institution of Great
Britain, W. L. Bragg Papers (RI MS WLB 86). 9C. Pattern of ellipses drawn to indicate stereographic projection.
From Bragg, 1913, figure 4 (see note 31).
Bragg subsequently called this the method of stereographic projection because the 2-D pattern of
spots is reconstructed as a 3-D system of cones pointing back into the photographic plate to atomic points in
intersecting panes of the crystal. This reconstruction turns a pattern of dots on the plate into a representation
of a section of a 3-D structure of X-ray beams created by physical processes, diffraction of the X-rays by the
crystal lattice and subsequent interference between diffracted rays. He went on to develop a method of
measuring the intensity of the deflected rays producing each spot, adapting an optical spectrometer that
deflected X-rays to ionize gas in a chamber connected to an electroscope. This action was measured as
electrostatic charge, quantifying the ionisation produced by a specific ray. This established a relationship
between the beam intensity and the angle of deflection.33 The X-ray spectroscope substitutes a precise
measure of intensity for Bragg’s estimated spot magnitude. This precision enabled him to calculate distances
between planes and from these, relative positions of atoms in the crystal planes.34
Science often aims to be able to make measurements that can test predictions. Yet here, as is often
the case, the physical meaning of the measured quantities is given by a simplified, visualized process model
of how an X-ray interacts with and is dispersed by a crystal lattice. This model is based on the original
analogy between a lattice and a diffraction grating but the full explanation combines several distinct models:
geometrical constructs which define crystal types (according to type of symmetry they display), geometrical
constructs which describe how the X-rays will behave in the crystal, mathematical models of the interference
of deflected rays and further geometrical constructs which relate the phenomenology (patterns of spots and
sizes of spots expressed as measured intensities) to the diffraction grating model, via the model of
interference. Given this complexity it is not surprising that Bragg devised a simple way of demonstrating
these relationships. He simulated the optical process postulated by the theory using polished, reflecting
rods arranged symmetrically around but diverging from a central axis (see figure 10b).35 Light reflected by
the rods is deflected to forms shadow-patterns similar to those in the X-ray images.
–– Envisioning Explanations, page 12 of 16 ––
A
B
FIGURE 10. 10A: Bragg’s earliest diagram of the diffraction of X-rays by a crystal (From Bragg 1913, fig 2, see
note 31). 10B: The model demonstrating how diffraction can produce the elliptical patterns that correspond to
positions of spots on the X-ray photographs such as figure (From Bragg 1933, figure 19, see note 35).
Like the paleontologists’ shadow-projection method and Teutsch’s liver process simulation this
device makes it easy both to conduct and to understand complex transformations between 2-D patterns and
3-D models of processes that produce those patterns.
As he moved from an interpretation of the
phenomenology of von Laue’s X-rays toward a coherent explanation of key features of those patterns, Bragg
followed a similar play of moves between simple and complex visualizations as we saw in the other cases.
The greater accuracy of the X-ray spectrometer method allowed Bragg to narrow possible lattices down to
two or even one structure. Yet he continued to prefer the stereographic method because the data produced,
like the spots in the X-ray patterns, have physical meaning only in the context of the patterns.36
Conclusion
David Hockney recently revived the thesis that the graphic arts owe much to imaging techniques developed
in other contexts.37 By demonstrating how light rays reconstruct on the retina an exact (albeit inverted)
replica of an external scene the camera obscura became far more than an imaging technology. It has
dominated science and philosophy as an explanation of how the external world is known to the minds eye,38
an epistemological engine which placed the viridicality of mental images beyond question for several
centuries.39 Chemists modelling complex molecules quote da Vinci’s observation that "the eye … is the chief
means whereby the understanding may --- appreciate the works of nature".40 The optical metaphor of the
camera obscura still lends credibility to processes of visualizing hidden structures as diverse as of crystals,
visceral organs and extinct arthropods. However it fails as an explanation of how ‘nature’ gets into ‘the
mind’s eye’ because it bypasses both the active dynamic of perception and the many objects, procedures and
technologies that we use to image our experience and what we imagine. Attending to the function of
different kinds of visual object shows how these techniques relate visual, verbal and other aspects of the
problem of envisioning explanations. Thus, projecting shadows of 3-D fossil models to verify informal
‘mental’ projections made in constructing 3-D sectional model from patterns (figures 3A and 3B) has its
counterpart in Bragg’s physical illustration of the analogy between light and X-rays (figure 11). These
cognitive artefacts make some aspects of visual thinking accessible to us.
The way science is practiced should reflect our cognitive make up just as it reflects our social,
political and other propensities. By comparing cases we can identify some features of human reasoning that
–– Envisioning Explanations, page 13 of 16 ––
combines visual with non-visual elements. These include images such as photographs, visualizations of
phenomena (diagrams, plots, graphs) and visual representations of theories about the phenomena (3-D
structures such as block diagrams, physical stacks of plots, and physical models). All are visual, but each is a
different type of thing. A diagrammatic abstraction from a photograph (whether of a fossil or an X-ray) is a
move from a barely interpreted source to an informative construct (see figure 4). This ‘move’ is motivated
by the desire to understand. It is informed by expert (often tacit) knowledge about indicative features which
need to be preserved or highlighted in the diagram. Block diagrams and process models cannot be purely
visual as the photos are because – as the use of labels, legends, or index numbers indicates -- they introduce
and seek to integrate other kinds of knowledge in order to produce an explanation. Examples are Bragg’s
analogy between X-ray diffraction by a 3-D crystal and optical diffraction by a 2-D grating and, in
paleontology, knowledge about the anatomy and physiology of known life forms.
I have shown that visual inference itself has a patterned character. This suggests another similarity
between art and science. Sometimes in art and usually in science, imaging methods have a particular
purpose: to produce an integrated image from spatially or temporally differentiated observations.
Perspective drawing, also based on a theory of rays, is a well-known case of a two-way exchange of ideas
and techniques between art and science. The drawing frame is a device that supports a technique for
ordering discrete acts of perception (looking through the drawing frame) into acts of depiction (marking
paper or canvas) integrating them into a single image.41 Picasso’s experiments with photography involved
making photographic plates of his paintings at various stages of completion. There is evidence that he
experimented with these plates by projecting light through stacks of them, thereby creating the multi-faceted
images that characterise early cubist art.42 The methods of structuring perception in our scientific examples
guide visual imagination rather than the artist’s hand yet they, too discipline perception in order to convey
understanding. My examples show that integrating many images into structural and process models
overcomes a spatial or temporal discontinuity and sometimes both together. This suggests that the ability to
move between two, three and four dimensions is what structures the play of interpretation and explanation
in many scientific fields.
Notes and Literature cited
1
A. E. POPHAM, ed., The Drawings of Leonardo da Vinci, London: The Reprint Society, 3 vols., 1946, vol. 3,
drawings on pages 279-281 and 290-296; B. FORD, Images of Science. A History of scientific Illustration, 1992,
London: British Library, J. DARIUS, Beyond Vision: One hundred historic scientific Photographs, 1984, Oxford:
Oxford University Press, D. M. KNIGHT, Natural Science Books in English, 1600-1900, 1972, London: Batsford;
H. ROBIN, The Scientific Image from cave to computer, 1992, New York: H. Abrams.
2
www.ed.ac.uk/iash/vkconf.html; see also A. M. BARRY, Visual Intelligence: perception, image and
manipulation in visual communication, 1997, Albany, NY: SUNY Press.
3
R. FORTEY, The Earth: an Intimate History, 2004, London: Harper Collins.
4
J. V. FIELD and F.A.J.L. James, eds., Science and the Visual, 1998, a special issue of the British Journal for the
History of Science, 31, R. MAZOLINI, ed., Non-verbal Communication in Science prior to 1900, 1993, Florence: L.
S. Olschki, M. LYNCH and S. WOOLGAR, eds., Representation in Scientific Practice, 1990, Cambridge, MA.:
MIT Press,
5
C. JONES and P. GALISON, eds., Picturing Science: Producing Art, 1998, New York and London: Routledge.
6
P. GALISON, Image and Logic. A material Culture of Microphysics, 1997, Chicago: Chicago University Press.
–– Envisioning Explanations, page 14 of 16 ––
7
Quoted in M. KELLY, ‘Whitehall’s Superconductor’, Times Higher Education Supplement, September 29, 1995,
p. 6. Similarly, Philip Ball remarks that “much of the scientist’s art lies in figuring out what to include and
what to exclude in a model, and this ability allows science to make useful predictions without getting
bogged down by intractable details”, P. BALL, The Self-Made Tapestry: Pattern Formation in Nature, p. 14, 1999,
Oxford: Oxford University Press.
8
A. CHALMERS, What is this Thing called Science?, 1978, Milton Keynes: The Open University Press, D.
GILLIES, Artificial Intelligence and Scientific Method, 1996 Oxford: Oxford University Press.
9
P. JOHNSON-LAIRD, The Computer and the Mind, 1988, London: Fontana, P. CARRUTHERS, P. STICH,
and M. SIEGAL, eds., The cognitive Basis of Science, 2002, Cambridge: Cambridge University Press.
10
Mithen, S. ‘Human evolution and the cognitive basis of science’ in P. CARRUTHERS, P. STICH, and M.
SIEGAL, eds., The cognitive Basis of Science, pp. 23-40, 2002, Cambridge: Cambridge University Press.
11
M. KEMP, Visualizations: the Nature book of Art and Science, p. 1, 2000, Oxford: Oxford University Press.
12
R. GIERE, Explaining Science: a Cognitive Approach, 1988, Chicago: Chicago University Press.
13
F. CRICK and J. WATSON, ‘Molecular Structure of Nucleic Acids’, Nature, 25 April 1953, 737-739.
14
T. NICKLES, ‘Justification and Discovery’, in D. GOODING, T. PINCH and S. SCHAFFER, eds., The Uses of
Experiment, pp. 299-333, 1989, Cambridge: Cambridge University Press.
15
D. GOODING, ‘Magnetic curves and the magnetic field: experimentation and representation in the history
of a theory’, pp. 183-223 in GOODING, PINCH and SCHAFFER, eds. (see note 14).
16
D. GOODING, Experiment and the Making of Meaning, Dordrecht: Kluwer Academic.
17
T. MARTIN, ed., Faraday’s Diary, entry for March 1832, vol. 1, p. 425, 1932-36, 7 volumes, London: G. Bell.
18
D. L. BRUTON, ‘The arthropod Sidneyia inexpectans, Middle Cambrian, Burgess Shale, British Columbia’,
Phil. Trans. Roy. Soc. B 295: 619-56; D. BRIGGS and S. CONWAY MORRIS, ‘New Burgess shale fossil sites
reveal middle Cambrian faunal complex’ 1983, Science, 222: 163-167; S. CONWAY MORRIS, ‘Middle
Cambrian polychaetes from the Burgess shale of British Columbia’, 1979, Phil. Trans. Roy. Soc. Lond. B285,
227-274; D.L. BRUTON and H. B. WHITTINGTON, ‘Emeraldella and Leanchoilia, two arthropods from the
Burgess Shale, British Columbia’, Phil. Trans. Roy. Soc, B 3000: 553-585, H. B. WHITTINGTON, 1985, The
Burgess Shale, New Haven, Conn., Yale University Press.
19
D. BRIGGS and S. H. WILLIAMS, ‘The restoration of flattened fossils’, 1981, Lethaia, 14. 157-164, D.
BRIGGS and D. COLLINS, 1988, ‘A middle Cambrian Chelicerate from Mount Stephen, British Columbia’,
Paleontology, 31: 779-798, Whittington, The Burgess Shale (see note 18). On the camera lucida see Hammond, J.
H. and Austin, J., The Camera Lucida in Art and Science, 1987, Bristol: Institute of Physics/Adam Hilger.
20
BRIGGS and WILLIAMS, p. 157, 1981 (see note 19).
21
Conway Morris, quoted in S. GOULD, p. 92, 1989, Wonderful Life. The Burgess Shale and the Nature of
History, London: Penguin. Gould 1989, p. 92)
22
H. WHITTINGTON, esp. figures 4.40, 4.50, 1985 (see note 18).
23
In practice, of course, researchers exchange sketches and verbal descriptions throughout.
24
G. GOLDSCHMIDT, ‘The dialectics of sketching’, 1991, Creativity Research Journal, 4, 123-143.
25
BRIGGS and WILLIAMS (see note 19), J. H. DOVETON, ‘Numerical methods for the reconstruction of
fossil material in three dimensions’, 1979, Geological Magazine, 116: 215-226.
26
The ‘convergent’ or integrating function of such images has been noted by historians: D. GOODING,
‘Mathematics and Method in Faraday’s Experiments’, 1992, Physis, XXIX, 121-147, M. TRUMPLER,
‘Converging Images: Techniques of Intervention and Forms of Representation of Sodium-Channel Proteins
–– Envisioning Explanations, page 15 of 16 ––
in Nerve Cell Membranes’, 1997, Jour. Hist. Biol., 30: 55-89, R. TWENEY, ‘Stopping Time: Faraday and the
Scientific Creation of Perceptual Order’, 1992, Physis, XXIX, 149-164,
27
28
P. BALL, The Self-Made Tapestry (see note 7).
H. TEUTSCH, ‘Modular design of the liver of the rat’, in G. MALCOLM, ed., Multidisciplinary Studies of
Visual Representations and Intrepretations, Elsevier, pp. 000-000 [ in press ]
29
H. TEUTSCH, ‘Regionality of Glucose-6-Phosphate Hydrolysis in the Liver Lobule of the Rat’, Hepatology,
1988, 8: 311-317; H. TEUTSCH, S. ALTEMUS, S. GERLACH-ARBEITER, and T. KYANDER-TEUTSCH,
‘Distribution of 3-Hydroxybutyrate Dehydrogenase in Primary Lobules of Rat Liver’, Journal of
Histochemistry and cytochemistry, 1992, 40: 213-219,
30
H. TEUTSCH, D. SHUERFELD, and E. GROEZINGER, ‘Three-Dimensional Reconstruction of
Parenchymal Units in the Liver of the Rat’, Hepatology, 1999, 29, 494-505.
31
W. L. BRAGG, ‘The Diffraction of short electromagnetic Waves by a Crystal’, Proc. Camb. Phil. Soc., 1913,
17: 43-57.
32
The situation is far more complex than his diagrams suggest. See W. L. BRAGG, ‘X-rays and Crystals’, p.
9, Jour. Roentgen Soc., 1914, X: 1-12.
33
W. L. BRAGG, ‘The Structure of some Crystals as indicated by their Diffraction of X-rays’, Proc. Roy. Soc.,
1913, A 89, 248-277, W. L. BRAGG, ‘The Analysis of Crystals by the X-ray Spectrometer’, Proc. Roy. Soc.,
1914, A 89, 468-489.
34
W. H. BRAGG and W. L. BRAGG, ‘The Reflection of X-rays by Crystals’, Proc. Roy. Soc, 1913, A 88, 428-438.
35
W. L. BRAGG, The Crystalline State: a General Survey, figure 19, p. 25, vol. 1, 1933, London: G. Bell.
36
W. L. BRAGG, The Crystalline State, p. 30 (see note 35).,
37
D. HOCKNEY, Secret Knowledge: Rediscovering the Lost Techniques of the Old Masters, 2001, London: Thames
and Hudson.
38
S. ALPERS, The Art of Describing, pp. 46-48, 1983, Chicago: Chicago University Press; M. BAXANDALL,
Patterns of Intention, 1985, New Haven, Conn.: Yale University Press.
39
D. IHDE, ‘Epistemology Engines’, Nature, 6 July 2000, 406: 21.
40
A. OLSON and D. GOODSELL, ‘Visualizing biological Molecules’, Scientific American, November 1992, pp.
44-51.
41
Alpers, The Art of Describing, p. 43 (see note 38). On the significance of printing and imaging technologies
for science see Ziman, J. Public Knowledge: an essay concerning the social dimension of science, 1968, Cambridge:
Cambridge University Press, pages 45-48.
42
A. MILLER, Einstein, Picasso: Space, Time and the Beauty that Causes Havoc, pp. 119-120, 136-145, 2001, New
York: Basic Books.
David C. Gooding
Science Studies Centre
Department of Psychology, University of Bath
Bath BA2 7AY, United Kingdom
[email protected]
–– Envisioning Explanations, page 16 of 16 ––